A skill that strips redundant, narrating, and self-explanatory comments so the code speaks for itself.
Note
This project is backed by the Dimwit Pledge.
Most comments an agent writes restate the line above them and rot the moment the code changes. This skill deletes those, keeps the ones carrying information the code genuinely cannot, and upgrades doc blocks that only echo the signature.
/no-comments- "clean up the comments in this file"
- "too many comments"
It also applies while writing or editing code in a repo that follows the convention.
- Deletes restatements, section banners, step narration, agent chatter, and commented-out code
- Keeps magic-number rationale, non-obvious "why", workarounds with tickets, gotchas, license headers, and anything a tool reads (
# noqa,// @ts-expect-error, shebangs, pragmas) - Rewrites signature-echo doc blocks into something a caller would be surprised by
- Prefers a better name or an extracted function over a comment explaining the mess
- Reports what was kept and what was rewritten — the interesting part of the diff
Works across JS/TS, Python, Go, Rust, Java, C/C++, shell, SQL, YAML, HTML, and CSS.
For every comment: would a competent reader learn anything from this that the code doesn't already tell them? If not, it goes.